NoSE: Schema Design for NoSQL Applications
26
NoSE: Schema Design for NoSQL Applications Michael J. Mior, Kenneth Salem, Ashraf Aboulnaga, Rui Liu
-
Upload
michael-mior -
Category
Technology
-
view
106 -
download
3
Transcript of NoSE: Schema Design for NoSQL Applications
NoSE: Schema Design forNoSQL Applications
Michael J. Mior, Kenneth Salem, Ashraf Aboulnaga, Rui Liu
NoSE
● NoSQL App Development
● Problem Formulation
● NoSE Design and Implementation
● Evaluation
NoSQL
● Eventually consistent, horizontally scalable, flexible schema
● Many different types of NoSQL databases○ Document stores○ Key-value stores○ Graph databases○ …○ Extensible record stores
Extensible Record Store Data Model
CREATE COLUMNFAMILY "ReservationsByGuest"(
"GuestID" uuid, "ResID" uuid,
"ResStartDate" timestamp,
"RoomID" uuid, PRIMARY KEY(("GuestID"),
"ResStartDate", "ResID", "RoomID")
);
Partitioning key
Clustering key
Database Application Development
1. Define application requirements
2. Decide on a data model for the target system
3. Implement the application according to the model
a. Database access
b. Application logic
Database Application Development
1. Define application requirements
2. Decide on a data model for the target system
3. Implement the application according to the model
a. Database access
b. Application logic
}NoSE
Schema Design Best Practices
Source: http://www.ebaytechblog.com/2012/07/16/cassandra-data-modeling-best-practices-part-1/
Schema Design Best PracticesModel column families around query patterns
Source: http://www.ebaytechblog.com/2012/07/16/cassandra-data-modeling-best-practices-part-1/
Schema Design Best PracticesModel column families around query patterns
But start your design with entities and relationships, if you can
Source: http://www.ebaytechblog.com/2012/07/16/cassandra-data-modeling-best-practices-part-1/
Schema Design Best PracticesModel column families around query patterns
But start your design with entities and relationships, if you can
De-normalize and duplicate for read performance
Source: http://www.ebaytechblog.com/2012/07/16/cassandra-data-modeling-best-practices-part-1/
Schema Design Best PracticesModel column families around query patterns
But start your design with entities and relationships, if you can
De-normalize and duplicate for read performance
But don’t de-normalize if you don’t need to
Source: http://www.ebaytechblog.com/2012/07/16/cassandra-data-modeling-best-practices-part-1/
Schema Design Best PracticesModel column families around query patterns
But start your design with entities and relationships, if you can
De-normalize and duplicate for read performance
But don’t de-normalize if you don’t need to
Leverage wide rows for ordering, grouping, and filtering
Source: http://www.ebaytechblog.com/2012/07/16/cassandra-data-modeling-best-practices-part-1/
Schema Design Best PracticesModel column families around query patterns
But start your design with entities and relationships, if you can
De-normalize and duplicate for read performance
But don’t de-normalize if you don’t need to
Leverage wide rows for ordering, grouping, and filtering
But don’t go too wideSource: http://www.ebaytechblog.com/2012/07/16/cassandra-data-modeling-best-practices-part-1/
Schema Design ExampleFor a given guest, return the cities that guest has stayed in
CREATE COLUMNFAMILY "CitiesByGuest" ("GuestID" uuid,
"City" text, PRIMARY KEY(("GuestID"), "City"));
CREATE COLUMNFAMILY "HotelsByGuest" ("GuestID" uuid,
"HotelID" uuid, PRIMARY KEY(("GuestID"), "HotelID"));
CREATE COLUMNFAMILY "HotelsByID" ("HotelID" uuid,
"HotelCity" text, PRIMARY KEY(("HotelID"), "HotelCity"));
NoSE Overview
Input Output
Conceptual
schema
Workload
Selected column
families
Query
implementation
plans
NoSE
Application Conceptual Model
Hotel
HotelIDHotelNameHotelPhoneHotelAddressHotelCityHotelStateHotelZip
Room
RoomIDRoomNumberRoomRateRoomFloor
Reservation
ResIDResStartDateResEndDate
Guest
GuestIDGuestNameGuestEmail
Point of Interest
POIIDPOINamePOIDescription
Amenity
AmenityIDAmenityDescription
Application WorkloadFor a given guest, return the cities that guest has stayed in
SELECT Hotel.HotelCity FROM Hotel.Room.Reservation.Guest
WHERE Guest.GuestID = ?
Hotel
HotelIDHotelCity
Room
RoomID
Reservation
ResID
Guest
GuestID
NoSE ArchitectureNoSE
Input OutputCandidate
Enumeration
Query Planning
Schema
Optimization
Plan
Recommendation
Conceptual
schema
Workload
Selected column
families
Query
implementation plans
Query Planning ExampleSELECT Name FROM Hotel WHERE Hotel.State = ‘NY’ AND
Hotel.Reservation.Room.Guest.GuestID = ? ORDER BY Name
GuestID↓
RoomID
RoomID↓
HotelID
HotelID↓
Name, State
Name
State
Schema Optimization
Construct a linear program to optimize execution time
Cost of using column family j to answer query i
Use of column family j for query i in the final plan
Presence of column family j in final schema
Size of column family j
Schema Optimization
Add constraints to ensure each query has a valid plan
Minimize the cost
Ensure column families used are present
Limit maximum storage space
Updates● Updates make denormalization more expensive
● Add statements to update conceptual entities
● New column families are added to support updates
● Costs for updates are added to the linear program
Evaluation
● Application defined by the RUBiS online auction benchmark
● Generate a schema and query plans recommended by NoSE
● Two schemas for comparison
○ Normalized (as much as possible)
○ Expert-selected
Evaluation - Schema Performance
Conclusion
● NoSE automates schema design for NoSQL applications
● Conforms to best practices without requiring expertise
● Schemas are better than those produced manually with an average of 1.8x and up to 125x performance improvement
Questions?
git.io/nose-icde
